EBOOK

REDSHIFT VS SNOWFLAKE
AN IN-DEPTH COMPARISON

AMAZON REDSHIFT SNOWFLAKE

Table of Contents 

Introduction 1 
Primary Use-Cases 2 
Architectural Difference 3 
Performance Differences 8 
Pricing Models 10 
Scalability 12 
Unique Features 14 
Bonus - How to Perform ETL to Redshift and Snowflake​? 16 

   
  
 
1 

Introduction 

 
Redshift

Amazon Redshift is an enterprise-level, petabyte scale, columnar, and 

fully managed data warehousing service from AWS. Its massive parallel 
processing and columnar compression make it one of the most efficient 
data warehousing services. It also supports standard SQL and is fast 
enough when compared to any traditional data warehouse. It also 
provides a quick option to load massive data sets.   

Snowflake

Snowflake is a simple, affordable, and high-performance data 

warehousing service on cloud built using AWS. It stores data in controlled 
columnar fashion. It provides broad support for standard SQL 
queries(update, delete, and joins). It was also built keeping in mind the 
challenges faced by conventional data warehousing systems. The 
management cost and effort is almost zero in this solution as there is no 
infrastructure to manage. It automatically handles security, optimization, 
infrastructure, etc. 
  
 
2 

Architectural Difference 

Redshift 

● Integrations

Amazon Redshift can be integrated with various ETL tools like Hevo, BI 
reporting like Power BI, and other analytics tools. Redshift follows 
industry-standard PostgreSQL hence most existing SQL client 
applications would work with least changes. 

   
  
 
3 

● Connections 

Amazon Redshift communicates with applications by using PostgreSQL 

JDBC and ODBC drivers. 

● Clusters 

The core component of the data warehouse is a cluster on Redshift. A 

cluster can have one or more compute nodes. The various nodes in 
Redshift cluster are following: 

1. Leader Node 

The leader node interacts with client programs and does all the 
communication with compute nodes. It communicates steps to obtain a 
certain result in the most efficient way, assigning data storage to all to 
compute nodes. It does not store any data and acts as a leader instructing 
all the compute nodes for the actions. 

2. Compute Nodes 

The leader node compiles code for the request and assigns the code to 
individual compute nodes. Now, all the compute nodes will execute the 
compiled code and send results back to the leader. 

Each compute node has its own CPU, memory, and disk storage, which 
are configured by the node type from AWS console login or CLI. 

Node Slices 

A compute node is made up of slices (partitions). Each slice has a portion 

of the node's memory and some disk space, where it processes the 
workload assigned to the node. The number of slices per node is defined 
by the node size of the cluster. 

   
  
 
4 

Internal network 

Redshift uses complex protocols (VPC) to provide highly secure and 

high-speed network communication between leader and compute nodes 
without hampering the performance. 

● Databases 

A cluster contains one or more databases. User data is stored on the 

compute nodes. The SQL client requests the leader node, which in turn 
asks query execution with the compute nodes in a systematic way. 
 
Amazon Redshift is an RDBMS, therefore it’s compatible with most of the 
RDBMS applications and it is customized for high-performance analysis 
and reporting/KPI’s of large datasets. 
 
Amazon Redshift resides on PostgreSQL 8.0.2. However, Amazon 
Redshift and PostgreSQL have few important differences that need to be 
accounted as you develop your data warehouse applications. 
  
 
5 

Snowflake 

Snowflake is a combination of shared disk database and shared-nothing 

database architectures. It fits for both structured and semi-structured 
data. As a feature of shared-disk architectures Snowflake has central 
data storage for prolonged data that can be accessed by all compute 
nodes in the warehouse. While as a feature of shared-nothing 
architectures, Snowflake queries are performed in parallel which is 
termed as MPP (Massively Parallel Processing). Snowflake uses 
micro-partitions to securely and efficiently store data. When loaded into 
Snowflake, data is automatically split into micro-partitions, and metadata 
is extracted to enable efficient query processing. This makes data 
management simple and the management cost comes to zero which is 
not in the case of Redshift. It also holds performance and scale-out 
benefits. 

   
  
 
6 

Database Storage 

When data gets loaded into Snowflake tables, the data is stored in the 
compressed and columnar format in the most optimized way. Snowflake 
uses Amazon Web Services S3 (Simple Storage Service) cloud storage 
for the same purpose. 

Snowflake manages almost all the admin and management aspects of 
how this data is stored in S3 — the size of a file, its structure, columnar 
compression, metadata definition of data storage. The data objects stored 
in S3 is not visible to customers. They can only be accessed through SQL 
query operations. 

Query Processing

All the query part is performed in the processing layer. Queries are 
processed using virtual warehouses. Virtual warehouse acts as an 
independent cluster allocated with separate workload as per our 
requirements. It uses AWS EC2 for achieving this purpose. It is the most 
prime feature of Snowflake as compared to Redshift which lacks such 
mechanism. 

Cloud Services

It is a layer for a collection of all the managed services that coordinate 

tasks across Snowflake architecture. Snowflake supports multiple ways 
of connecting to most of the services. 

A web-based user interface and command line clients (e.g. SnowSQL) 

both are capable of managing and using Snowflake. 

 
  
 
7 

Primary/Foreign Keys Constraints 

Redshift 
 
Primary keys and foreign key constraints are just for information. They are 
not mandatory in Redshift. However, primary keys and foreign keys are 
used to design an effective query plan by the query engine. Hence, it is a 
good practice to declare them. The query planner uses these relationships 
but it assumes that all keys in Amazon Redshift tables are valid as loaded. 
So, we need to show extra care with integrity constraints. If the 
application allows invalid keys, few queries could return spiked results. 
Amazon Redshift enforces NOT NULL column constraints. Data 
distribution, workload management of queries, data partition, configuring 
nodes, clusters, table sorting, and S3 are some unique feature to store 
and access the data in the most efficient way. 
 
Snowflake 

Snowflake also supports defining constraints but does not enforce them, 
except for NOT NULL as in the case of Redshift. Snowflake supports 
constraints on permanent, transient, and temporary tables. Constraints 
can be defined on any number of columns of any data types. For 
Snowflake Time Travel (data recovery), when previous versions of a table 
are recovered, the current version of the constraints on the table is used 
because the history of metadata is not stored on Snowflake. It is a zero 
management data warehousing service as data distribution, workloads, 
configuring nodes, backups, and most of the tasks related to managing 
and storing data are either managed by Snowflake or are a matter of few 
clicks. Snowflake focuses on analyzing the data more rather than 
managing them. We can create many virtual warehouses and configure 
them as per need. It is very cost effective and easy to create.   
  
 
8 

Performance Differences 
Redshift 

Redshift has various ways to get high-performance parallel queries. 

Experts believe that it results in a speedup of 8 times on long-running 
queries over PostgreSQL, MySQL, and SQL Server. 
 
● Workload management:​ Database admins can control query 
queues where queries can be provided more priority over any other 
ETL jobs as per our requirement. 
● Data compression:​ Individual columns in Redshift are stored 
separately. We can define the compression type while creating the 
table. It also helps in high throughput while transferring data across 
the cluster. 
● Query optimizer: I​ t is intelligently designed for massively parallel 
processing as per the trend of modern data warehousing service. 
 

Snowflake   

Snowflake is a relational columnar-store cluster warehousing solution 

similar to Redshift supporting MPP. 

● Virtual computation warehouses:​ Snowflake provides the 

capability of creating virtual warehouses for each of your 
independent tasks. For example, a reporting query can hit a virtual 
warehouse 1 and KPI query can hit virtual warehouse 2 as they are 
independently querying the data the performance remains the 
same. Also, ETL jobs can run on slower and less expensive 
warehouse and business related queries can run on 
high-performance warehouse during business hours. You can easily 
scale up, down, or pause compute power. Also, you only pay when 
you query. 
● Data retention:​ Snowflake has a time travel feature which can help 
you easily revisit the historical data anytime between the last ninety 
days. Redshift can be configured for auto backup in S3. 
  
 
9 

● Automatic tuning efforts​: Snowflake self-tunes the performance of 

the system as you use it. It even takes care of scaling and resizing 
as per demand. A very little hands-on admin approach is required 
as it manages optimization related tasks on its own. Thus, you 
barely need a database admin to perform the mentioned tasks. In 
Redshift, a database admin is required.

Note: ​For a cluster that runs 24 hours a day Redshift is the best option. 
Whereas for reporting queries and when ETL is only done when required 
then Snowflake is a better option as you are only charged when you 
query the data warehouse. 
 

   
  
 
10 

Pricing Models 
Redshift 

Redshift operates on two kinds of pricing model as mentioned below: 

 
● On-demand Pricing - Pay at an hourly rate. 
● Reserved Instance Pricing - 1 or 3-year contract and it is 75% 
cheaper than the on-demand model. 

Redshift charges are based on the number of hours and number of nodes. 
The pricing starts at $0.25 per-hour for 160GB data. Redshift lets you 
choose the hardware specifications as per your requirement. It helps you 
find how much storage and throughput you get from the money invested. 

 
 
Source:​ A
​ WS Redshift Pricing

   
  
 
11 

Snowflake 

Snowflake pricing largely depends upon the usage pattern. It charges an 
hourly rate for each of the virtual warehouses created. Data storage is 
decoupled so it is charged separately as $0.20 per TB per month. It offers 
7 different types of the warehouse. The X-small is the smallest which is 
charged at $2 per hour. Snowflake offers dynamic pricing model which 
means clusters will shut down when not in use and automatically start 
when in use. They also can be resized on the fly depending upon the 
workload thus saves more money. 
 

Source: S
​ nowflake Manual 

Choosing the Right Cluster

Selecting the right cluster depends on your usage patterns. If the cluster 
is up and running 24 hours a day (due to ETL or reporting) Redshift is the 
better option. If the ETL runs ones in a week and only querying of data is 
required as per demand Snowflake is the option. 

   
  
 
12 

Scalability 
Redshift 

Let’s consider if you are trying to load 1TB of data for the below 
instances. Data load speeds are proportional to the number of nodes 
defined in the cluster as shown by the findings below: 
 
● A single node XL instance will take close to 16 hours. 
● A multi-node XL instance of two nodes will take close to 9 hours. 
● A multi-node 8XL instance of two nodes will take close to 1.5 
hours. 

Querying the data 

A query will run faster when there is a number of nodes but the 
performance does not rise linearly. Redshift clusters are optimized for 
multiple node clusters supporting MPP in the best possible way. 

Resizing 

Redshift offers to resize, closing, launching of the cluster by a simple API 

call or by few mouse clicks from AWS console. The clusters can be 
upscaled or vice versa with few minutes of downtime. 

   
  
 
13 

Snowflake  

As Snowflake is easy to use and accessible on almost any scale for all the 
users and applications deployed on the cloud. It manages storage, 
compute, and metadata separately. Billions of rows of data can be 
queried by concurrent users sitting anywhere. Storage and compute can 
be scaled up or down independently and the metadata service will 
automatically scale up and down as per the requirement. 

In the environment, shutting down database operations for overnight is 

not required as Snowflake does this of its own. We can create 
independent clusters on the fly and assign it to the users based on priority 
and requirement. Thus you can have different users, different compute 
capacity but all pointing to the same data lake.  

   
  
 
14 

Unique Features 
Redshift 

● Automatic Columnar Compression: I​ t provides better performance 

at lower costs.  
● Elastic MapReduce Access​: If you have data stored in EMR Data 
then it can be copied from an Elastic MapReduce cluster to a 
Redshift cluster. 
● Concurrency: Y ​ ou can configure a maximum of 50 simultaneous 
queries across all user queues (Workload management).We can 
increase the concurrency to get better query performance for some 
long running queries.   
● Max Result Set Size: T ​ he cursor counts and result set sizes can be 
configured.However, read the documentation carefully before 
proceeding with this step. 
● Resizing Indicator: Y​ ou can monitor the progress of cluster resizing 
task in the AWS Redshift console. 
   
  
 
15 

Snowflake 
 
● Full SQL database:​ It supports DDL, DML, analytical functions, 
transactions, and complex joins. 
● Variety of data​: Snowflake ingests almost all kinds of data, either 
from traditional sources or machine-generated sources without 
tradeoffs. Snowflake supports both structured and semi-structured 
data like JSON and Avro. 
● No management: S ​ nowflake is a data warehouse as a service 
running in the cloud and thus there is no infrastructure to manage 
or knobs to turn. Snowflake automatically handles infrastructure 
requirement, optimization of queries or tables, data distribution, 
availability, and data security. 
● Performance: S ​ nowflake processes reports and KPIs at very high 
speed because of the columnar database engine. 
● Broad ecosystem: ​Snowflake integrates with almost all kind of 
tools near to its ecosystem like Hevo, Redshift, BigQuery. The 
different custom connectors include ODBC, JDBC, Javascript, 
Python, Spark, R, and Node.js.   
  
 
16 

How to Perform ETL to Redshift and Snowflake? 

AWS Redshift and Snowflake are high performing databases. However, 

migrating data from sources to Amazon Redshift and Snowflake involves 
multiple complex stages and can be a cumbersome experience. 

If you want to load any data easily into Redshift and Snowflake without 
any hassle, you can try out ​Hevo​. Hevo automates the flow of data from 
various sources to Amazon Redshift and Snowflake in real time and at 
zero data loss. In addition to migrating data, you can also build 
aggregates and joins on Redshift and Snowflake to create materialized 
views that enable faster query processing. 

 
Looking for a simple and reliable way to bring Data from
Any Source to AWS Redshift and Snowflake?

TRY HEVO
SIGN UP FOR FREE TRIAL